How to perform user segmentation and achieve refined operations?

The author of this article mainly shares the application methods and simple practical cases of the RFM model in the field of user stratification and refined operations . enjoy~

In the field of user operation , there is a term called RFM. I believe many people have seen it, but perhaps they feel it is too professional and do not go into it in depth; or some articles involve data processing methods such as Python, which makes people discouraged; or because this term is closely related to e-commerce , non-e-commerce operation partners choose to give up learning it.

Today I will mainly share the application methods and simple practical cases of the RFM model in the field of user stratification and refined operations.

The RFM model is one level higher than the user value model, and there are two directions in the user value model:

• One is to build a life cycle model based on the user life cycle , that is, the time and the user's growth path within the product.
• The other is to build an RFM model based on key user behaviors. This time we will only talk about the RFM model in the user value model.

What are the benefits of the RFM model? After we have determined the RFM model, we can decide which users to add the prefix "Dear VIP user" when sending text messages, and which users to add the prefix "Long time no see". It can also help companies determine which users have unusual movements and whether there are signs of user loss, thereby increasing corresponding operational actions. Its great uses are shown below.

You can search the Internet for more information about RFM. I won’t go into details here. Let me first explain the meaning of the three letters:

• R: Recency , which refers to the time since the user's last consumption. Of course, the shorter the time since the last consumption, the better. It is more valuable for us and more likely to reach them effectively.
• F: Consumption frequency (frequency) , the frequency of consumption of users within the product within a period of time. The focus is on our definition of a period of time.
• M: consumption amount (monetary) , representing the value contribution of the user.

At first, 5 levels were defined in each direction of R, F, and M. 5 5 5 = 125 user categories. This is too complicated for most operations and products. You don’t need to understand such historical issues as why it is divided into 5 levels. Now we have defined each direction of R, F, and M as high and low. We find the median of R, F, and M. R = the most recent consumption. If it is higher than the median, it is high, and if it is lower than the median, it is low. This way, there are 2 2 2 = 8 user classifications, as shown in the following figure:

Therefore, if we can find out which of the above 8 categories the users in the product belong to, we can formulate targeted operation strategies.

Before we get into the specific RFM setup, I want to emphasize once again that the RFM model is not only applicable to the e-commerce field, but also to other fields as well. As long as we find the data fields related to R, F, and M, define the fields well, and prove that these fields are the most critical dimensions that affect the current business progress:

• R: Last login time, last post time, last investment time, last viewing time
• F: Number of views, posts, and comments
• M: Deposit amount, reward amount, number of comments, number of likes

The above are definitions of R, F, and M in other fields, and the specific definitions should be evaluated based on actual business conditions. For example, if you are the operations manager of Douban , and you find that the overall traffic of Douban's book section has dropped by 10% in the past week, and at the same time, the number of shared posts under the literature book category has dropped by 5% month-on-month, you want to analyze the reasons. You can select the corresponding R, F, and M fields respectively: number of logins, number of posts, and number of interactions.

• The overall traffic of the book section has decreased, which can be understood as a decrease in the activity of this section. You can look at the number of logins in the past week.
• The number of shared posts in the literature book category dropped by 5%. You can look at the number of posts in the past week.
• At the same time, when traffic decreases, we can see whether the decrease in traffic is caused by a relative decrease in the quality of posts, which leads to a decrease in user interactions (comments, favorites, etc.).

Below, I will take a set of 10,000 pieces of data from a shopping guide platform as an example, and show you how to use the simplest method to build a user RFM model in this set of data and find out these 8 categories of users.

The steps to build the RFM model are as follows:

• Capture the original data in three dimensions: R, F, and M
• Define the evaluation model and median of R, F, and M
• Perform data processing to obtain the values ​​of R, F, and M
• Stratify users by referring to the evaluation model and median value
• Specify operation strategies for users at different levels

(1) Capture the original data under the three dimensions of R, F, and M. I captured the most recent consumption time, consumption frequency, and consumption amount. As mentioned above, when doing business analysis, you can select the data fields of R, F, and M according to the actual situation. The following table shows 13 of the 10,000 pieces of data:

(2) I processed the most recent consumption time, consumption frequency, and consumption amount of 10,000 pieces of data using a percentage trend chart . Taking consumption frequency as an example, as shown in the following figure:

From the chart, you can see that among the 10,000 data pieces, there are several obvious gaps in consumption frequency, namely: consumption 1 time, consumption 2-5 times, consumption 6-11 times, consumption 12-17 times, and consumption more than 18 times. Therefore, I divide the F value into 5 levels, F=1=consumption 1 time, F=2=consumption 2-5 times, F=3=consumption 6-11 times, F=4=12-17 times, F=5=18 times or more.

Similarly, using the method shown in the figure above, I found the data ranges corresponding to the R value and M value of 5 respectively. R=1=2 days, R=2=3-8 days, R=3=9-14 days, R=4=15-22 days, R=5=23 days or more; M=1=600 yuan, M=2=601-3800 yuan, M=3=3801-6200 yuan, M=4=6201-10000 yuan, M=5=10001-15000 yuan.

We get the classification standards under the three RFM data indicators:

(R value is an inverse value, the larger the R value, the lower the user value; F value is a positive value, the larger the F value, the higher the user value; M value is a positive value, the larger the M value, the greater the user value.)

(3) Calculate the R, F, and M values ​​corresponding to the most recent consumption time, consumption frequency, and consumption amount for each piece of data:

The calculation method in the above figure is relatively simple. We write the if statement in Excel:

Cell E2 = if (B2>23,5,if(B2>15,4,if(B2>9,3,if(B2>3,2)))))

explain:

• If B2 is greater than 23, then the R value corresponding to the A1 user = 5, otherwise go to the next if judgment;
• If B2 is greater than 15, then the R value corresponding to the A1 user = 4, otherwise go to the next if judgment;
• If B2 is greater than 9, the R value corresponding to the A1 user = 3, otherwise it goes to the next if judgment;
• If B2 is greater than 3, the R value corresponding to user A1 = 2, otherwise go to the next if count.

The F value is calculated in the same way as the M value.

(4) Calculate the average value of R, F, and M. Everyone should know how to do this. Simply sum them up and divide by the number of items. R(ave)=2.9, F(ave)=1.8, M(ave)=2.7

(5) Compare the R value, F value, and M value of each user in the 10,000 data points with the average value . If the value is higher than the average value, it is marked as high, and if the value is lower than the average value, it is marked as low:

To compare the high and low values, use a simple if statement:

H2=if(E2<2.9, "low", "high"), the calculation of F value and M value is similar.

(6) Compare the R, F, and M values ​​of each user with the median value to obtain the user category table: The R value, F value, and M value of each user are compared with the median value to determine whether they are high or low, and then determine which category the user belongs to among the 8 categories of users in the RFM model mentioned above. Here we need to use a simple if statement to make a judgment. Let's take user A1 as an example to determine the user category to which user A1 belongs:

K2=IF(AND(H2=”High”,I2=”High”,J2=”High”),”Important Value Users”, IF(AND(H2=”High”,I2=”Low”,J2=”High”),”Important Development Users”,IF(AND(H2=”Low”,I2=”High”,J2=”High”),”Important Retained Users”, IF(AND(H2=”Low”,I2=”Low”,J2=”High”),”Important Retained Users”,IF(AND(H2=”High”,I2=”High”,J2=”Low”),”General Value Users”, IF(AND(H2=”High”,I2=”Low”,J2=”Low”),”General Development Users”,IF(AND(H2=”Low”,I2=”High”,J2=”Low”),”General Retained Users”,”General Retained Users”)))))))

At the same time, we click on "Conditional Formatting" in Excel and set the text with the word "high" to green and the text with the word "low" to red, which makes it easier for us to identify.

So far, we have obtained a complete and refined stratification of users under these 10,000 data items. Next, we can make corresponding operational strategies based on the stratification results and carry out specific implementation work.

(7) Formulate operational strategies based on user stratification results

When formulating operational strategies, we must consider both the proportion of each type of user in the product and the actual business logic of the product. Taking the user stratification of a shopping guide platform as an example, the following strategy is formulated:

When some friends formulate strategies, they directly throw out a bunch of "plans" that cannot be called strategies. For example, for "important development users", the strategy I give is "increase frequency". All means around increasing frequency can be tried, instead of formulating plans such as: sending push , issuing coupons, making phone calls, etc. These are all operational means supported by strategies. The strategy itself must be extendable and replicable.

In addition to formulating operational strategies based on user categories as mentioned above, we can also analyze the R value distribution, F value distribution, and M value distribution in the 10,000 data items. Based on the distribution of the three values ​​and the comparison with the median value, we can perform overall operations based on the dimensions of the most recent consumption time, consumption frequency, and consumption amount to improve the overall activity of users on the site, the overall process, and increase GMV.

Overall, the RFM model is not very difficult, but there are some points to note:

1. When capturing raw data, it is important to select key data indicators for analysis based on actual business, rather than the stereotyped time of most recent consumption, frequency of consumption, and amount of consumption . The Douban example was also given above. In the Douban case, R = number of logins, F = number of posts, and M = number of interactions.
2. When defining the evaluation model of R value, F value, and M value to separate data intervals, it is not always necessary to use the method described in this article to look at the overall trend and find obvious gaps. You can also use scatter plots, pivot tables, proportion charts, etc. to make judgments. At the same time, in addition to discovering gaps through data, we can discover critical points based on our own business and the industry average. For example, for taxi-hailing apps like Didi and Yidao, the usage frequency is definitely relatively high on weekdays. Therefore, if we analyze Didi's business, the five levels of F-value consumption frequency can be based on actual business, with every five days as a level, to analyze the business performance in the past 30 days. For example, F=1=within 5 days, F=2=6-10 days, F=3=11-15 days, F=4=16-20 days, F=5=21-30 days. Then compare the extracted consumption frequency of each user with these five levels to determine the F value of each user.
3. The simplest way to calculate the median is the average value calculation method mentioned in this article. In addition to the average, there is also the 80/20 rule, which states that 20% of users generate 80% of revenue, so this critical point can be used as the object of comparison for each user's R, F, and M. For more complex businesses, you can seek help from programmers and use the Means clustering algorithm to obtain accurate data.
4. In addition to selecting three core business indicators for cross-analysis as mentioned in this article, sometimes we may need to analyze four or five indicators at the same time, or only need to analyze two indicators. Therefore, there is no need to use the calculation method in this article rigidly. You need to be flexible and adaptable . I will not give examples here.
5. Ultimately, we still have to return to operations. Therefore, the formulation of operational strategies for different user strata must be based on actual conditions . After the operational strategies are formulated, specific implementation work should be carried out in combination with the company's existing resources and means.

It is recommended that you try out the contents of this article.